The lowest GPU prices in the market, published.
SaladCloud gives you on-demand access to thousands of RTX-class GPUs, with live supply visible alongside published per-second pricing. Four priority tiers let you balance availability, reliability, and cost. vCPU and RAM are included with every GPU.
Reserved capacity is quoted separately under Salad Dedicated.
Every GPU class, right now.
GPUs online and schedulable at medium and high priority. Higher priority can take over lower-priority workloads, so the high figure is never smaller than the medium figure. Machines already reserved on contract are not included in these counts.
| GPU class | Relative supply | Medium priority | High priority |
|---|
Dark green is medium priority, light green is high. Medium is part of high, so matching counts show only the dark bar.
Estimate a deployment.
Choose a GPU and priority tier, then compare the estimate with current capacity above. Availability varies by class, region, and priority.
Final cost is determined by the seconds each instance spends in the running state. Allocation, download, and cold start are free. Higher priority tiers are protected from preemption by other workloads; Lowest is the lowest cost and the most interruptible.
Per GPU, per hour. vCPU and RAM included.
Every GPU class ships with vCPU and RAM at no additional charge. Choose the priority tier per container group; you can mix tiers across a deployment.
| GPU | VRAM | High | Medium | Low | Lowest | $/mo (Lowest) |
|---|---|---|---|---|---|---|
| RTX 5090 | 32 GB | $0.500 | $0.417 | $0.333 | $0.250 | $182.50 |
| RTX 4090 | 24 GB | $0.330 | $0.273 | $0.217 | $0.160 | $116.80 |
| RTX 5080 | 16 GB | $0.260 | $0.223 | $0.187 | $0.150 | $109.50 |
| RTX 5090 Laptop | 24 GB | $0.235 | $0.190 | $0.145 | $0.100 | $73.00 |
| RTX 4080 | 16 GB | $0.230 | $0.190 | $0.150 | $0.110 | $80.30 |
| RTX 5070 Ti | 16 GB | $0.225 | $0.183 | $0.142 | $0.100 | $73.00 |
| RTX 4070 Ti Super | 16 GB | $0.210 | $0.170 | $0.130 | $0.090 | $65.70 |
| RTX 4070 Ti | 12 GB | $0.200 | $0.160 | $0.120 | $0.080 | $58.40 |
| RTX 3090 Ti | 24 GB | $0.190 | $0.160 | $0.130 | $0.100 | $73.00 |
| RTX 3090 | 24 GB | $0.170 | $0.143 | $0.117 | $0.090 | $65.70 |
| RTX 5070 | 12 GB | $0.165 | $0.137 | $0.108 | $0.080 | $58.40 |
| RTX 3080 Ti | 12 GB | $0.155 | $0.130 | $0.105 | $0.080 | $58.40 |
| RTX 5060 Ti | 16 GB | $0.150 | $0.123 | $0.097 | $0.070 | $51.10 |
| RTX 4070 | 12 GB | $0.150 | $0.123 | $0.097 | $0.070 | $51.10 |
| RTX 4060 Ti | 16 GB | $0.140 | $0.120 | $0.100 | $0.080 | $58.40 |
| RTX 3080 | 10 GB | $0.140 | $0.113 | $0.087 | $0.060 | $43.80 |
| RTX 5060 | 8 GB | $0.110 | $0.095 | $0.080 | $0.065 | $47.45 |
Additional classes (RTX 3060, RTX 3070, RTX 3070 Ti, and others) are available in the portal from $0.015 per GPU-hour at Lowest priority. GPU classes and live availability: API reference.
Pay for the reliability you need.
High
Never preempted by other workloads. Still subject to node disconnection; SCE reallocates automatically.
Medium and Low
May be preempted by higher-priority groups. Minimum run before reallocation is roughly 30 minutes.
Lowest (Batch)
The lowest-cost tier. Ideal for retryable batch work and stateless inference behind a queue. CPU-only groups run here.
CPU-only container groups.
Billed at $0.005 per vCPU-hour and $0.001 per GB of RAM per hour. Common configurations:
General purpose
| RAM | vCPU | $/hr |
|---|---|---|
| 1 GB | 1 | $0.006 |
| 2 GB | 2 | $0.012 |
| 4 GB | 2 | $0.014 |
| 8 GB | 4 | $0.028 |
| 16 GB | 8 | $0.056 |
| 16 GB | 16 | $0.096 |
CPU-optimized
2:1 memory to vCPU for consistent, dedicated-core performance.
| RAM | vCPU | $/hr |
|---|---|---|
| 2 GB | 1 | $0.007 |
| 4 GB | 2 | $0.014 |
| 8 GB | 4 | $0.028 |
| 16 GB | 8 | $0.056 |
Memory-optimized
Extra memory per vCPU to avoid out-of-memory errors and swapping.
| RAM | vCPU | $/hr |
|---|---|---|
| 8 GB | 1 | $0.013 |
| 16 GB | 2 | $0.026 |
| 24 GB | 3 | $0.039 |
| 32 GB | 4 | $0.052 |
Volume, reserved, and private.
Volume discounts
Deploying more than 100 GPUs, long-running jobs, or committed spend? Sales will price it below list.
Talk to sales →Salad Dedicated
Reserved RTX PRO 6000 Blackwell servers and VMs on contract, quoted per allocation and term.
Learn more →Private Cloud
Licensed per deployment on your own hardware. Marginal cost of a token: electricity.
Learn more →Simple and transparent. No surprises.
Do I pay for cold boot time?
No. You pay only for time the hardware is running your application. Allocation, image download, and container start are free.
When do charges start and stop?
Billing is per second while an instance is in the running state. It stops the moment the instance stops, is reallocated, or is preempted.
Do you offer volume discounts?
Yes, for deployments above 100 GPUs, long-running jobs, and committed contracts. Fill out the sales form and we'll price your workload.
Is there a free trial?
No. SaladCloud and the Transcription API are pay-as-you-go with no minimum commitment, so you can start with a small, low-cost test before scaling.
How can prices be this low?
Community Cloud runs on GPUs that already exist in homes and small businesses worldwide. Owners are paid for idle time, and a competitive marketplace keeps prices near the cost of electricity rather than the cost of a data center.